Fix two bugs in new S1 AFT and pattern likelihood calculations

[pdeperio]

1. Event index i was overwritten by a loop index over PMTs, so compute_gof and s1_aft_map.get_value were always using the position of event number 247. Now fixed by using more explicit variable names.

2. The argument pmt_selection in compute_gof should be an array of booleans of size apc[self.tpc_channels]. Not sure if what's done in BuildInteractions.py is correct, but here it was crashing at

   
   pmts_bottom = np.setdiff1d(self.tpc_channels, confused_s1_channels)
   pmts_bottom[self.pax_config['DEFAULT']['channels_top']] = 0

dask.async.IndexError: index 118 is out of bounds for axis 1 with size 118

since ```pmts_bottom``` is no longer the full PMT array after removing ```confused_s1_channels```. Now I just explicitly keep a full PMT array ```is_pmt_in```, similarly to [```TopPatternFit.py```](https://github.com/XENON1T/pax/blob/v6.8.3/pax/plugins/posrec/TopPatternFit.py) to keep track of which PMTs to include in the likelihood.

@Requested reviewers, please double check this implementation and implications.

[coderdj]

Fine but prefer we test that updates don't break things before merging them. These variables now useless int the production we just did.